Information Technology Reference
In-Depth Information
the information from other slices is also taken into consideration. The initial
results after this phase are promising. The comparision of the contours obtained
for 10 (1 chosen slice and 10 phases of heart cycle) 2D images in segmentation and
improvement phase is presented in Fig. 7 and Fig. 8 (the reslut of segmentation
phase was the initial contour in the improvement phase and additional potential
sources were added to increase contour flexibility). A slight improvement can
be observed especially in recall value of the right ventricle. However, the visual
effect (Fig. 6c, Fig 6d) reveals that better improvement technique still should be
searched for. The reason behind it is a fact that only visual information (even
if additional slices are used) is considered and an expert possesses additional
medical knowledge that allows to draw a proper contour. It can be used in the
presented framework because any information can be used in energy function. For
example, one can consider the conscious expert knowledge (e.g. the relationship
between shapes of both ventricles such as information that the interventricular
septum should have a constant thickness) as well as some knowledge that is not
conscious or is hard to express directly using mathematical formula and that
can be obtained using machine learning techniques (e.g. neural network trained
to estimate the proper localization of the contour in the given slice taking into
account the image information from other slices). These aspects are under further
investigation.
5
Summary
In this paper three-phase approach for segmentation of left and right heart ven-
tricle was presented. The approach in question imitates the conscious top-down
human process of segmentation using adaptive potential active contour (APAC)
technique. The obtained results reveal that proposed methodology is well suited
to the described problem. The preprocessing and segmentation phases allow to
achive very good approximate localization of heart ventricles. However, a precise
review of expert results requires additional semantical knowledge. The proposed
improvement phase uses only visual information and, thus, the the improvement
is only slight. It shows that in that phase other formulations of energy function,
especially those that are able to use unconscious or hard to express medical ex-
pert knowledge, should be used, which will be further investigated by the authors
of this work.
Acknowledgement
This work has been supported by the Ministry of Science and Higher Educa-
tion, Republic of Poland, under project number N 519007 32/0978; decision no.
0978/T02/2007/32.
Authors would like to express their gratitude to Mr Cyprian Wolski, MD,
from the Department of Radiology and Diagnostic Imaging of Barlicki University
Hospital in Lodz for making heart images available and sharing his medical
knowledge.
